It is not a stupid question.

Whilst the goal would be to produce those materials in reality and later down the line even mass produce them, there are a few hurdles to overcome.

You can arrange atoms one by one with current technology but as you might imagine that doesn’t scale, so you need a chemical process that produces the desired output, and that can be tricky, to put it mildly.

So, since the efforts and resourses to create a new way to sythetize new materials are huge, the benefit of AI predicted models come in the form of estimating which possible materials have the properties that we are seeking, so less money goes to waste developing ways to create them.

But we are very far away from a Star Trek replicator, sadly.